Event contracts.

A Kalshi contract pays a dollar if something happens. Bookie asks whether its price is wrong often enough to pay the fee, holds every rule to settled markets it never saw, and bets only what survives — which so far is nothing, so it has bet nothing. Augur sits beside it and cannot bet at all: it says how likely, from a source better placed to know than the crowd, and keeps score against the price. This is a stream by design and not yet by revenue, and this page will say when that changes. The sibling desk is equities.

bookie · kalshi event contracts Armed · nothing to bet

$ bookie bet --stake 5 --max-bets 10

hourly at :05 · armed 2026-09-20 · the gate is on and there is no flag to remove it

studied45,562 settled daily-high weather markets

rules5 held to markets they never saw

survive0

nearest+0.21% a bet, p = 0.81 against p < 0.01

bets0

waiting3 rules that lean on Augur’s forecasts — held-out trial not yet run

The hypothesis was the old one about favourites: that a contract priced at 90¢ wins more often than nine times in ten. On Kalshi’s weather markets it does not, or not by more than the fee — favourites there are priced about right. So Bookie wakes every hour, reads its own journal, finds nothing that survived, and exits before it opens a connection. Its key can trade and cannot move money. Economics, index, crypto and sports markets are still being fetched; each rule tested on them costs a trial, and the threshold tightens with every one. The next question is already registered: not whether a 90¢ contract is worth 90¢, but whether it is worth it when a forecaster says otherwise. Three such rules were chosen from early markets only and wait for their one look at the late ones.

Fig. 01 — Bookie, whose usual answer is that the price is about right.

augur · forecasts Forecasting · holds no key

$ augur weather run && augur score

hourly at :50, fifteen minutes before bookie decides · live since 2026-09-21 · figures as of then

rebuilt356,600 forecasts, each from what was knowable that hour

checked~39,000 settled markets reproduced from the climate reports, 100%

said 20%came in 20.4% — calibrated, which is not the same as useful

vs crowdbuying YES on its say-so lost 14–25% a bet, in-sample

lanesweather · CPI · the Fed · web search, which bookie cannot see

Augur turns the National Weather Service’s model blend into the probability each temperature bracket pays, using only the forecast that existed at the time and an error model that learns only from days it could have known. It does the same for inflation from the Cleveland Fed’s daily nowcast, and restates what fed funds futures imply about the next meeting. The first thing it found was against itself: where the blend and the market disagree, the market is usually right — the crowd has the blend too, and the radar. One narrow case held up on early data, buying NO soon after a market opens when a confident forecast says the bracket will miss, and that is the hypothesis Bookie will test once. A fourth lane asks Claude, with web search, about markets nobody publishes a forecast for; it costs money per question, cannot be backtested, and is scored but deliberately not shown to Bookie. Augur has no signing code and runs as its own user. The most a web page can do by talking to it is make it wrong, and being wrong is something the gate already knows how to price.

Fig. 02 — Augur, which says how likely and may not say how much.